Skip to content

Video Extension

Extend existing videos with AI-generated content.

Overview

Video extension allows you to continue an existing video with new AI-generated content. The AI analyzes the last frames and generates continuation based on your prompt.

Requirements

Critical Requirements

Requirement Details
Frame Rate Source video MUST be 24fps
Model veo-2.0, veo-2.0-exp, or veo-3.1-generate-preview
Source Local file or GCS URI (NOT external URLs)

Why 24fps?

The Veo API only accepts 24fps videos for extension. Videos with different frame rates will fail with "fps mismatch" error.

Basic Usage

# Extend a local video
python video_gen.py "The cat runs faster" \
  --extend-video previous_video.mp4 \
  --model veo-2.0

# Extend with GCS storage
python video_gen.py "Continue the scene" \
  --extend-video previous_video.mp4 \
  --storage-uri gs://my-bucket/videos \
  --model veo-2.0

Source Options

Local File

python video_gen.py "Continue scene" \
  --extend-video /path/to/video.mp4

The script will:

  1. Check if video is 24fps
  2. Upload to GCS if --storage-uri provided
  3. Use base64 encoding if no GCS

GCS URI

python video_gen.py "Continue scene" \
  --extend-video gs://my-bucket/videos/source.mp4

Direct GCS URI is faster - no upload needed.

With Reference Images

Combine extension with reference images:

python video_gen.py "The person continues walking" \
  --extend-video previous.mp4 \
  --reference-image avatar.png:asset \
  --model veo-2.0

With Last Frame

Specify where the video should end:

python video_gen.py "Transform to sunset" \
  --extend-video daytime.mp4 \
  --last-frame sunset.jpg \
  --model veo-2.0

Workflow: Creating Long Videos

Chain multiple extensions to create longer videos:

# Step 1: Generate initial video
python video_gen.py "A person walking" --preset quality
# Output: video_001.mp4

# Step 2: Extend first video
python video_gen.py "The person starts running" \
  --extend-video video_001.mp4 --model veo-2.0
# Output: video_002.mp4

# Step 3: Extend again
python video_gen.py "The person jumps over obstacle" \
  --extend-video video_002.mp4 --model veo-2.0
# Output: video_003.mp4

# Combine with ffmpeg
ffmpeg -i "concat:video_001.mp4|video_002.mp4|video_003.mp4" -c copy final.mp4

Common Issues

"fps mismatch"

Problem: Source video is not 24fps

Solution: Convert to 24fps using ffmpeg:

ffmpeg -i input.mp4 -r 24 -c:v libx264 output_24fps.mp4

"video is empty"

Problem: External URL used (not supported)

Solution: Download video first, then use local file:

# Download video
curl -o local_video.mp4 "https://example.com/video.mp4"

# Convert to 24fps if needed
ffmpeg -i local_video.mp4 -r 24 local_24fps.mp4

# Use local file
python video_gen.py "Continue" --extend-video local_24fps.mp4

"Model does not support video_extension"

Problem: Using unsupported model (e.g., veo-3.0)

Solution: Use supported models:

python video_gen.py "Continue" \
  --extend-video input.mp4 \
  --model veo-3.1-generate-preview  # or veo-2.0

Supported models for video extension: - veo-3.1-generate-preview ✅ - veo-2.0 ✅ - veo-2.0-exp ✅

Smart Defaults (New in v2.25.1)

When extending a video, settings are automatically inherited from the source video's metadata:

How It Works

# Original video was created with:
# - model: veo-2.0
# - resolution: 1080p
# - storage-uri: gs://my-bucket/videos

# Just specify the prompt - settings are inherited
python video_gen.py "Continue the scene" --extend-video video_20260201_123456_0.mp4

# Output:
# ℹ Found source metadata for inheritance
# ℹ Inherited model from metadata: veo-2.0
# ℹ Inherited resolution from metadata: 1080p
# ℹ Inherited storage_uri from metadata: gs://my-bucket/videos

Priority Chain

Settings are resolved in this order:

  1. CLI Arguments (highest) - What you type
  2. Source Metadata - From the source video
  3. config.json - Your project settings
  4. Code Defaults (lowest) - Built-in fallbacks

Override When Needed

You can always override inherited values:

# Override model (ignore metadata)
python video_gen.py "Continue" \
  --extend-video video.mp4 \
  --model veo-2.0-exp

# Override resolution
python video_gen.py "Continue" \
  --extend-video video.mp4 \
  --resolution 720p

Chained Extensions

Smart Defaults work perfectly for chained extensions. Each video inherits from its predecessor: 

video_001.mp4 → video_002.mp4 → video_003.mp4
# All use same model, resolution, storage settings

Auto-Upload Feature

When --storage-uri is provided, local files are automatically uploaded:

python video_gen.py "Continue scene" \
  --extend-video local_video.mp4 \
  --storage-uri gs://my-bucket/uploads

Flow: 

Local file → Upload to GCS → Use gcsUri in API → Generate → Download

Best Practices

Do

  • ✅ Use Veo-generated videos (always 24fps)
  • ✅ Use --storage-uri for reliable uploads
  • ✅ Use supported models (veo-2.0, veo-2.0-exp, or veo-3.1-generate-preview)
  • ✅ Keep prompts consistent with source video

Don't

  • ❌ Use external URLs directly
  • ❌ Use unsupported models (e.g., veo-3.0)
  • ❌ Use videos with non-24fps frame rate
  • ❌ Create drastic scene changes

Planned Features (v2.25)

Future version will support:

# Planned: Auto-download and convert external URLs
python video_gen.py "Continue" \
  --extend-video-from-url "https://example.com/video.mp4"

This will:

  1. Download video from URL
  2. Convert to 24fps using ffmpeg
  3. Upload to GCS
  4. Proceed with extension

Learn More